The Difference That Makes A Difference In Predicting Future Business Success

Predictive modeling is all about leveraging historical business data to predict the likelihood of future success.

But hold on a moment: if this assertion is correct, its very premise seems to go against that now-infamous – yet ubiquitous – cautionary phrase found on advertisements for investment opportunities anywhere they appear in print:


Past performance is not necessarily indicative of future results.

This line of thinking begs the question: how can what we know about the past predict what will happen in the future?

The short answer is that predictive modeling does not actually predict the future; nothing can. However, what proficient predictive modeling does quite well is that it can tell us with an amazing degree of precision how likely a certain possible future is.

To paraphrase the great 20th century systems thinker and polymath Gregory Bateson, the accuracy of a predictive model depends upon knowing which differences make a difference.

Bateson’s Definition Of Information

Bateson, whose book “Steps To An Ecology Of Mind” served as the inspiration for our company name, famously defined information as a “difference which makes a difference.” In its more abstract interpretations, his definition has broad implications in fields ranging from the theory of logical types to learning theory and from information theory to the study of schizophrenia – even to the so-called “mind-body problem” in philosophy.

In the context of predictive modeling, we can loosely apply Bateson’s definition of information in this sense:

The data that we analyze for possible inclusion in a predictive model can only be viewed as information when it makes a difference in improving the predictivity of the model (i.e., how accurately the model can predict a given future business scenario).

Types Of Future Business Scenarios To Which Predictive Modeling Can Apply

There is a huge range of possible future business scenarios to which predictive modeling can be applied, helping business decision-makers to answer such question as:

a. To which households should we send direct mail – and which should we skip over entirely?
b. Where should we build our next store or branch?
c. To which customers should we try to upsell or bundle high-margin products?
d. In which neighborhoods should we implement a billboard, cable TV, or newspaper campaign?
e. Which companies should we aggressively pursue in order to entice them to relocate to our town or city?

A 30,000-Foot View Of The Predictive Modeling Process

Here is a high-level view of how our predictive modeling process works:

1. Define a business decision whose outcome matters to the business owners.
2. Determine a threshold for the outcome that signifies whether it is desirable or not, either as a binary result (e.g., Yes/No, Good/Bad, etc.) or on a continuum (e.g., sales revenue figures).
3. Collect data (i.e., variables) related to past instances of similar business decisions.
4. Pre-test which individual variables seem to correlate with said outcome and cull those variables from the initial variable list into a “finalist” list of variables.
5. Use artificial neural network software or other machine-based learning technique to train a new predictive model. The objective is to find the most ideal combination and weighting of each of the finalist variables in terms of how well it predicts said outcome with the highest-possible degree of accuracy.
6. Feed into the newly-trained predictive model a new set of data that pertains to analogous instances of the same type of business decision to be made in the future, asking the model to predict the likely outcome.

Modeling Helps Determine Which Differences Make A Difference

As per the above 30,000-foot description, our modeling process reflects Bateson’s “difference that makes a difference” at two points: #4 (pretesting) and #5 (model-building).
The pretesting phase helps us to determine which variables make a difference, while the model-building phase tells us how much each variable matters and in combination with which other variables.

This approach brings the abstract definition of a visionary like Bateson into the realm of practical business and economic decision-making.